Face recognition method based on semi-supervised training

Abstract

The invention relates to a face recognition method based on semi-supervised training, and belongs to the field of computer vision. The method includes: firstly, using a face recognition data set as labeled data, obtaining face pictures from the Internet to serve as label-free data, and obtaining training data through face detection and alignment for the labeled data and the label-free data; introducing a loss function based on a label-free picture, and carrying out semi-supervised training together with the loss function of the labeled picture; introducing a task balance factor alpha and a data balance factor beta to balance the relationship between a supervised task and an unsupervised task. Compared with other face recognition methods using label-free data, the method has the advantagesthat label-free pictures do not need to be clustered, the mode of using the label-free pictures is more efficient, and the performance of the model is better; according to the method, good performanceimprovement is achieved on a plurality of face recognition test sets, and good universality is achieved.

Description

technical field [0001] The invention relates to a face recognition method based on semi-supervised training, in particular to a face recognition method for semi-supervised training based on labeled pictures and unlabeled pictures, and belongs to the technical field of computer vision. Background technique [0002] With the development of deep learning, the accuracy of face recognition models has also been greatly improved. Face recognition technology has been widely used in intelligent security, financial payment, access control punching and other fields, and has extremely high commercial value. At present, the scale of face recognition data sets has exceeded tens of millions of pictures and 100,000 face categories. Further collecting more labeled data requires a lot of manpower and material resources. However, there is a large amount of unlabeled data in the Internet that has not been properly utilized. [0003] At present, researchers have used unlabeled data to optimiz...

AI Technical Summary

Problems solved by technology

Further collecting more labeled data requires a lot of manpower and material resources

However, there is a large amount of unlabeled data in the Internet and it has not been used reasonably.

[0003] At present, researchers have used unlabeled data to optimize face recognition models, but these methods all need to cluster unlabeled data, and the accuracy of the clustering algorithm itself is low, and ultra-large-scale clustering requires a large amount of memory with time

After clustering, the number of each category is inconsistent, resulting in serious category imbalance

Images